Doubao AI Agent Store Launches First 1,200 Enterprise Agents: Why Token Aggregation Platforms Are Becoming the Key to Affordable AI Scaling
In the summer of 2026, ByteDance’s latest AI moves revealed a much clearer strategic direction:
AI is moving from a conversational assistant into a full-scale enterprise productivity infrastructure.
On July 30, ByteDance CEO Liang Rubo announced an internal organizational adjustment:
The Feishu product team was integrated into the Doubao AI business unit, while the sales organization was transferred to Volcano Engine.
The move was significant.
One of ByteDance’s most important enterprise collaboration products was being deeply merged into its AI ecosystem.
Soon afterward, Doubao Enterprise Edition began internal testing among selected Feishu customers, integrating core workplace scenarios including:
Feishu Docs,
spreadsheets,
meetings,
enterprise communication,
collaborative workflows.
The message was clear:
ByteDance is transforming AI from a “conversation tool” into a business productivity platform.
From AI Chatbots to Enterprise Digital Workers
Another important signal appeared during the Volcano Engine FORCE conference on June 23.
ByteDance officially introduced Doubao’s paid membership system.
Together with enterprise AI platforms such as:
Coze,
HiAgent,
ArkClaw,
ByteDance is building a three-layer AI agent ecosystem:
Consumer AI Layer
Designed for individual users:
daily assistance,
information search,
personal productivity.
Enterprise Agent Platform Layer
Designed for organizations:
workflow automation,
business process optimization,
enterprise knowledge management.
Vertical AI Agent Layer
Designed for specialized scenarios:
finance,
legal,
coding,
data analysis,
industry applications.
Within this ecosystem, the first batch of enterprise AI agents launched in the agent marketplace reached approximately:
1,200 enterprise-level AI agents
covering critical business scenarios such as:
financial analysis,
data governance,
contract review,
software development,
enterprise operations.
The direction is becoming increasingly obvious:
AI agents are no longer designed merely to “talk.”
They are being built to complete real business tasks.
The Shift From “Human-Like Interaction” to “Business Execution”
The timeline is particularly interesting.
Around the same period when Doubao’s AI agent marketplace expanded, Doubao and Qwen adjusted some highly personalized conversational agent features.
The official explanation focused on product optimization.
However, the underlying industry logic is clear:
Entertainment-style AI agents create challenges:
high computing consumption,
uncertain monetization,
expensive safety management,
long-context memory costs.
Every user interaction requires:
model inference,
computing resources,
security evaluation,
memory management.
If user engagement increases without a stable business model, platform costs grow rapidly.
Therefore, the industry direction is changing:
AI platforms are moving from emotional interaction toward measurable productivity.
Doubao 2.1 Pro: AI Agents Enter the Engineering Workflow
The release of Doubao 2.1 Pro further confirms this strategy.
In a chip design RTL testing scenario:
Doubao 2.1 Pro reportedly operated continuously for nearly 18 hours, completing:
simulation,
testing,
synthesis verification,
multiple engineering iterations.
In benchmark evaluations including:
Terminal Bench 2.1,
SWE Pro,
OSWorld,
Doubao 2.1 Pro has entered the global top-tier AI agent performance category.
The key milestone is not simply model intelligence.
It is practical execution capability.
Once AI models cross the threshold of “usable,” the next challenge becomes:
How can enterprises afford large-scale AI agent deployment?
The answer leads to one critical resource:
Token.
Token Costs: The Hidden Challenge Behind Enterprise AI Expansion
The biggest difference between enterprise AI agents and consumer chatbots is not intelligence.
It is usage depth.
A traditional chatbot may answer one question.
An enterprise AI agent may need to:
collect business data,
analyze information,
compare historical records,
generate recommendations,
trigger automated workflows.
Every step consumes tokens.
Industry estimates suggest that AI agents may consume:
50 to 200 times more tokens than traditional chatbot applications when completing the same business objective.
This creates a new enterprise challenge:
AI capability is improving rapidly.
But AI operating costs are becoming harder to predict.
The Token Cost Trap: More Powerful AI Creates More Complex Bills
Doubao’s daily token usage has reportedly exceeded:
180 trillion tokens per day
Meanwhile, enterprise customers with annual token consumption exceeding one trillion tokens have continued increasing.
The problem is no longer whether enterprises can access AI.
The problem is:
Can enterprises understand and optimize where their AI budget goes?
Current AI infrastructure still has several limitations:
fragmented token supply,
fixed vendor pricing,
limited market competition,
inefficient resource allocation.
The industry needs a new economic layer.
That layer is:
The Token marketplace.
Unexhub: Turning Tokens Into a Tradable AI Resource
Unexhub identifies a fundamental gap in the AI ecosystem.
Its positioning is straightforward:
A global Token aggregation and distribution compute marketplace.
The platform connects:
Upstream AI suppliers:
DeepSeek,
Zhipu AI,
Doubao,
Gemini,
ChatGPT,
Claude,
GLM,
and other AI model providers.
Downstream users:
developers,
enterprises,
AI application builders.
The goal is to build liquidity infrastructure for global Token resources.
Instead of treating Token as a simple billing unit controlled by individual vendors:
Unexhub aims to transform Token into:
accessible resources,
dynamically priced commodities,
market-driven AI infrastructure.
Unexhub’s Three Mechanisms: Building the Economic Infrastructure Behind AI Agents
The emergence of enterprise AI agents creates a new challenge:
More intelligent AI requires more frequent model calls.
More workflows require more Token consumption.
More enterprise adoption requires a more efficient resource market.
Unexhub focuses on solving this infrastructure problem through three core mechanisms:
1. Instant Settlement: Reducing Financial Pressure Across the AI Supply Chain
Traditional AI resource distribution models create significant financial pressure for smaller AI providers.
The conventional process looks like this:
AI suppliers purchase computing resources or Token capacity in advance.
They provide API services to customers.
They wait for periodic payment settlement.
This creates several problems:
capital is locked in advance purchases;
cash-flow cycles become longer;
smaller providers face higher entry barriers;
innovative AI teams struggle to scale distribution.
Unexhub is introducing an instant settlement and clearing mechanism designed to improve liquidity throughout the Token ecosystem.
The concept is simple:
AI suppliers can recover value based on actual Token usage instead of waiting through long settlement cycles.
Traditional model:
Provide service first, wait for payment later.
Token marketplace model:
Provide resources, receive value settlement faster.
This transforms AI resources from a slow-moving inventory asset into a more liquid digital resource.
For AI suppliers, this means:
reduced financial pressure;
improved cash-flow efficiency;
greater ability to scale services.
2. Early Supplier Incentives: Lowering the Barrier to Enter the AI Marketplace
A healthy AI marketplace requires a diverse supplier ecosystem.
However, many emerging AI providers face challenges:
limited customer acquisition channels;
high distribution costs;
difficulty competing with major platforms.
To accelerate ecosystem growth, Unexhub plans to provide early-stage supplier support.
Qualified early suppliers can access:
0% Platform Commission During the Initial Ecosystem Stage
This means AI providers can:
reduce marketplace entry costs;
retain more revenue from API distribution;
focus resources on improving models and services.
For independent developers and smaller AI companies, this creates a lower-risk path to reach global users.
Instead of building an entire distribution network independently:
They can connect once and access a broader AI consumption market.
3. API Dynamic Bidding: Creating Market-Based Token Pricing
Today’s AI API market still mainly follows fixed pricing.
A model provider sets:
input price,
output price,
usage rules.
Users accept the pricing structure.
But AI resources are not static.
Demand changes.
Capacity changes.
Usage patterns change.
AI infrastructure increasingly behaves like:
electricity markets,
cloud computing resources,
financial markets.
It requires dynamic pricing.
Unexhub is exploring an API bidding mechanism where Token prices can respond to real-time market conditions.
Pricing can adjust according to:
supply availability;
demand intensity;
usage periods;
resource conditions.
Benefits for Developers and Enterprises
During lower-demand periods:
more Token capacity becomes available;
market prices become more competitive;
batch workloads can run at lower costs.
Examples:
data processing;
AI training experiments;
content generation;
large-scale analysis tasks.
Enterprises can choose the most economical execution window.
Benefits for AI Suppliers
During high-demand periods:
valuable capacity receives better pricing;
suppliers can maximize resource utilization;
previously idle computing resources can generate revenue.
The result:
Token begins developing its own “peak and off-peak pricing system.”
Just like electricity markets:
off-peak hours provide lower-cost access;
peak demand creates higher resource value.
Unused AI capacity is no longer wasted.
Scarce computing resources are no longer forced into inefficient fixed pricing.
Why Token Marketplaces Matter for Doubao’s 1,200 Enterprise AI Agents
The timing between Doubao’s enterprise agent expansion and Token marketplaces is particularly important.
The first 1,200 enterprise AI agents solve one problem:
“What AI workers can companies use?”
But another question remains:
“How can companies afford to operate these AI workers at scale?”
Every enterprise agent consumes Token resources.
A financial analysis agent requires:
data retrieval,
calculation,
reasoning,
report generation.
A coding agent requires:
code understanding,
testing,
debugging,
repeated execution.
Without efficient Token infrastructure:
Enterprise AI deployment can become economically difficult.
Unexhub addresses this challenge through:
Token Aggregation
Connecting multiple AI suppliers into one resource network.
Intelligent Routing
Selecting suitable AI models based on task requirements.
Dynamic Pricing
Allowing market competition to reduce unnecessary costs.
Real-Time Settlement
Improving supply-side liquidity.
Together, these mechanisms create a more sustainable economic model for enterprise AI adoption.
Engineer Perspective: AI Agent Marketplace + Token Marketplace = A New Development Paradigm
For developers, the combination of AI agent platforms and Token marketplaces is creating a new way to build applications.
1. AI Agents Are Becoming Ready-to-Use Digital Workers
Previously, enterprises often needed to build AI systems from scratch:
select models;
develop workflows;
create prompts;
integrate APIs;
maintain infrastructure.
Now, AI agent marketplaces are changing the process.
With thousands of enterprise-ready agents available:
Developers can:
select existing solutions;
customize workflows;
integrate faster;
focus on business innovation.
This is similar to the transformation from:
building private servers →
using cloud platforms →
using SaaS applications.
2. Token Costs Are Becoming Dynamic Instead of Fixed
Traditional AI spending:
Every API call follows the same price.
Future AI spending:
The best price depends on timing, supply, and routing strategy.
Through Token marketplaces:
non-urgent workloads can run during lower-cost periods;
high-priority workloads can access premium resources;
companies gain more control over AI budgets.
AI cost management becomes an optimization problem.
3. AI Computing Is Transforming From Heavy Infrastructure Into Flexible Resources
Traditional AI competition focused on:
GPU ownership;
data center scale;
hardware investment.
The next stage focuses on:
Token efficiency;
resource allocation;
intelligent scheduling;
market liquidity.
For example:
Doubao 2.1 Pro’s cost efficiency compared with some overseas flagship models demonstrates the growing competitiveness of domestic AI models.
When combined with Token marketplace infrastructure:
AI computing resources can move from:
expensive fixed assets
toward:
flexible, tradable digital resources.
The Future AI Economy: From Owning Compute to Optimizing Token Value
The AI industry is entering a new phase.
The question is no longer:
“How many GPUs does a company own?”
The more important question is:
“How much value can each Token generate?”
The future AI economy will depend on:
efficient resource distribution;
transparent pricing;
intelligent routing;
scalable AI access.
Unexhub: Building the Global Token Economy Infrastructure
Unexhub is creating a Token aggregation and distribution center designed to connect AI suppliers and users worldwide.
The platform brings together:
major AI model providers;
independent AI developers;
enterprise AI solution providers.
Through:
aggregation,
intelligent routing,
competitive pricing,
settlement infrastructure,
Unexhub aims to make AI resources more accessible and efficient.
For Developers and Enterprise Users
Users can benefit from:
✅ One unified gateway connecting multiple AI models.
✅ Pay based on actual Token consumption.
✅ Intelligent routing to match tasks with suitable models.
✅ Competitive marketplace pricing.
✅ Optimized access that can reduce AI usage costs significantly, with eligible scenarios potentially reaching around 30% of traditional market pricing.
For AI Suppliers
Suppliers can access:
✅ One-click platform integration.
✅ Global distribution opportunities.
✅ Revenue sharing based on API usage.
✅ Early ecosystem support with potential 0% platform commission benefits.
✅ Faster commercialization without building independent sales channels.
Conclusion: When Tokens Become a Market, AI Becomes More Accessible
Doubao’s enterprise AI agent ecosystem answers:
“What can AI do?”
Token marketplaces answer:
“How can AI scale economically?”
Together, they represent the next stage of AI commercialization.
AI is moving from:
models → agents → markets → AI economy
The future belongs not only to companies that build powerful models.
It belongs to companies that can make AI resources:
affordable,
accessible,
efficient,
scalable.
Unexhub is building the infrastructure that helps transform Token from a cost metric into a global AI economic resource.
豆包智能体商店首批 1200 个企业级智能体上线:Token 聚合分发中心正在成为 AI 低成本规模化调用的关键
2026 年夏天,字节跳动围绕 AI 生态的一系列战略调整,让行业看到了更加清晰的发展方向:
AI 正从一个聊天工具,逐渐演变为企业生产力基础设施。
7 月 30 日,字节跳动 CEO 梁汝波发布内部邮件,对组织架构进行调整:
飞书产品团队整体并入豆包业务体系,销售体系划归火山引擎。
这一变化意义重大。
作为字节跳动企业协作领域的重要产品,飞书正在与 AI 能力进行更深度融合。
随后,豆包企业版开始在部分飞书客户中展开测试,并原生集成:
飞书文档;
表格;
会议;
群聊;
企业知识管理;
等办公场景。
字节跳动的战略方向已经十分明确:
让 AI 不只是回答问题,而是真正参与企业业务流程。
从“像人聊天”到“帮助企业完成工作”
6 月 23 日,在火山引擎 FORCE 原动力大会上,字节跳动进一步释放 AI 商业化信号。
豆包正式推出付费会员体系。
与此同时,围绕:
扣子 Coze;
HiAgent;
方舟 ArkClaw;
等企业级智能体平台,字节正在构建一个完整的智能体生态。
这一生态可以理解为三个层次:
第一层:消费级 AI 助手
服务普通用户:
日常问答;
内容创作;
信息获取;
个人效率提升。
第二层:企业级智能体平台
服务企业:
工作流程自动化;
企业知识管理;
数据分析;
业务决策辅助。
第三层:行业垂直智能体
服务专业场景:
财务分析;
合同审核;
软件开发;
数据治理;
行业应用。
在这套体系中,豆包智能体商店首批上线约:
1200 个企业级智能体
覆盖多个企业核心场景:
财务分析;
数据管理;
合同审查;
代码开发;
企业运营。
这意味着:
AI 智能体正在从“展示能力”进入“创造业务价值”的阶段。
AI 竞争进入新阶段:重点不再是会不会回答,而是能不能完成任务
过去几年,大模型竞争主要围绕:
参数规模;
Benchmark 成绩;
推理能力。
但企业真正需要的不是一个更会聊天的模型。
而是:
一个能够持续执行任务的数字员工。
这也是为什么字节跳动开始弱化部分娱乐化、拟人化 AI 交互方向。
原因很简单:
聊天型 AI:
调用频率高;
算力消耗大;
商业闭环不稳定。
每一次用户交互,都意味着:
模型推理;
算力消耗;
安全审核;
上下文管理。
如果没有稳定商业模式,用户越活跃,平台成本压力越大。
因此,行业正在转向:
减少情绪价值,提升工作价值。
豆包 2.1 Pro:AI 开始进入真实工程流程
豆包 2.1 Pro 的发布进一步证明了这一趋势。
在芯片设计 RTL 测试场景中:
豆包 2.1 Pro 连续运行接近 18 小时,完成:
仿真;
测试;
综合检查;
多轮工程优化。
同时,在:
Terminal Bench 2.1;
SWE Pro;
OSWorld;
等智能体评测中,豆包 2.1 Pro 已进入全球领先模型行列。
这代表一个重要变化:
AI 已经跨过“能不能用”的阶段。
新的问题变成:
企业如何承担大规模 AI 智能体运行成本?
答案指向一个核心资源:
Token(词元)
企业 AI 最大挑战:Token 账单正在成为新的成本黑洞
企业级智能体与普通聊天机器人最大的区别,不只是能力。
更重要的是:
调用深度。
一个聊天机器人可能只需要一次回答。
但一个企业智能体可能需要完成:
获取业务数据;
分析历史信息;
计算关键指标;
生成决策建议;
触发后续流程。
每一步都消耗 Token。
业内测算:
完成同一个业务目标时,智能体消耗 Token 量可能达到传统 Bot 的:
50—200 倍
豆包大模型日均 Token 调用量已经达到:
180 万亿 Token
同时,年度 Token 消耗超过万亿规模的企业客户数量持续增长。
AI 使用规模正在爆发。
但企业面临的新问题是:
AI 越强大,成本为什么越难控制?
Token 成本困境:价格下降,但企业账单依然上涨
目前 AI 市场存在一个明显矛盾:
Token 单价越来越低。
但企业总支出仍不断增加。
原因:
企业调用规模增长速度超过价格下降速度。
例如:
模型价格下降 50%。
但如果企业调用量增长 10 倍:
最终成本仍然会上升。
因此,企业需要解决的不是:
“如何买最低价模型?”
而是:
“如何优化整个 Token 使用链路?”
包括:
从哪里购买 Token;
如何选择模型;
如何动态调度;
如何降低闲置成本。
这正是 Token 交易市场存在的意义。
Unexhub:让 Token 从计费单位变成可交易 AI 资源
Unexhub 正关注 AI 产业中的这一核心痛点。
平台定位:
全球 Token 聚合分发的算力交易中心
通过连接:
上游:
DeepSeek;
豆包;
智谱;
Gemini;
ChatGPT;
Claude;
GLM;
等 AI 模型供应商。
下游:
企业;
开发者;
AI 应用构建者。
Unexhub 希望建立一个更加开放的 Token 流通体系。
让 Token 从:
厂商制定价格的计费单位
转变为:
市场参与者自由交易的数字资源。
三大机制:重构 AI Token 经济模式
第一:Token 聚合分发,让企业一个入口连接多个 AI 模型
当前 AI 市场高度碎片化。
企业通常需要:
注册多个模型账号;
对接不同 API;
管理多个账单;
维护不同认证体系。
这增加了技术成本和运营复杂度。
Unexhub 通过 Token 聚合分发模式:
将不同 AI 供应商连接到统一平台。
企业无需重复管理多个供应体系。
通过统一接口:
即可调用不同 AI 模型。
结合智能路由能力:
系统可以根据任务需求匹配更适合的模型。
帮助企业:
降低开发成本;
提高调用效率;
优化 Token 消耗。
第二:即时结算,提升供应商资金流动效率
传统 AI 服务模式中:
供应商往往需要提前采购资源。
然后等待用户消费后才能回收资金。
这会导致:
资金占用;
回款周期长;
中小供应商进入门槛高。
Unexhub 正构建更加高效的结算机制。
供应商提供多少 Token 服务,即可根据实际调用获得对应收益。
简单理解:
过去:
卖出去以后等待结算。
未来:
服务产生价值,更快速完成收益回收。
为了鼓励更多 AI 服务商加入生态:
Unexhub 将针对早期入驻供应商提供阶段性扶持。
其中包括:
早期 0 佣金生态支持
帮助供应商:
降低平台进入成本;
提升收益空间;
更快速触达全球开发者和企业用户。
第三:API 动态竞价,让 Token 价格由市场决定
传统 API 模式:
供应商定价。
用户接受。
但未来 AI 资源更像:
电力;
云计算;
数据资源。
需要动态价格体系。
Unexhub 正探索:
API 动态竞价模式
让 Token 价格根据:
供需关系;
使用时间;
资源状态;
动态变化。
对于企业和开发者:
低峰时间:
Token 供应更充足;
调用成本更低;
批量任务更适合运行。
对于供应商:
高峰时期:
稀缺资源价值提升;
算力利用率提高;
收益更加合理。
未来:
Token 将像电力一样拥有:
“峰谷价格体系”
闲置算力不会浪费。
优质资源不会被低价出售。
豆包 1200 个智能体 + Token 市场:AI 商业化的重要组合
豆包智能体商店解决的问题:
企业可以使用哪些 AI 智能体?
而 Token 市场解决的问题:
企业如何低成本运行这些智能体?
两者结合,解决了 AI 商业化最大的鸿沟:
从“AI 能工作”到“AI 能规模化工作”。
工程师视角:智能体市场正在改变 AI 开发模式
第一,智能体正在成为标准化数字员工
过去:
企业需要从零搭建:
模型选择;
Prompt 设计;
工作流开发;
API 集成。
未来:
企业可以直接选择成熟智能体。
类似:
过去自建服务器。
后来使用云服务。
现在直接使用 SaaS。
AI 智能体也正在经历同样变化。
第二,Token 成本从固定费用变成动态优化问题
过去:
任何时间调用,价格一样。
未来:
根据:
时间;
供应;
路由;
选择最优成本。
企业 AI 预算将更加智能化。
第三,算力正在从重资产变成流动商品
过去竞争:
比谁拥有更多 GPU。
未来竞争:
比谁能够更高效利用 Token。
AI 算力正在从:
固定基础设施
转变为:
可交易、可调度、可优化的数字资源。
2026 年 AI 预算新逻辑:不是购买更多算力,而是提升 Token 价值
未来企业衡量 AI 投资,不再只是:
买多少 GPU;
部署多少模型;
建多少服务器。
而是:
每一个 Token 能创造多少业务价值。
Unexhub 正推动全球 Token 经济基础设施建设。
平台聚合:
大型 AI 模型供应商;
中小 AI 服务商;
独立开发者资源。
通过:
Token 聚合;
智能路由;
API 竞价;
即时结算;
帮助企业和开发者获得更加灵活、高效的 AI 使用方式。
加入 Unexhub,共建下一代 Token 经济生态
对于开发者和企业用户:
✅ 一个入口连接多种 AI 模型资源;
✅ 按实际 Token 使用量付费;
✅ 毫秒级智能路由匹配最佳模型;
✅ 通过市场竞争降低 AI 调用成本;
✅ 优化场景下,Token 使用成本最低可达到市场价格约 3 折。
对于 AI 供应商:
✅ 一键接入全球分发渠道;
✅ 无需独立建设销售体系;
✅ 根据 API 调用获得收益分成;
✅ 早期生态阶段享受 0 佣金扶持;
✅ 实现更快速商业变现。
结语:当 Token 成为市场,AI 才真正走向规模化
豆包智能体商店回答了:
“AI 可以做什么?”
Token 交易市场回答了:
“AI 如何被企业长期使用?”
未来 AI 产业的发展路径正在清晰:
模型 → 智能体 → Token 市场 → AI 经济生态
真正赢得未来的,不只是拥有强大模型的企业。
而是能够让 AI:
更低成本;
更高效率;
更大规模;
服务全球用户的基础设施。
Unexhub,正在推动 Token 从成本单位走向全球 AI 经济资源。